Upload
nguyenthuan
View
222
Download
3
Embed Size (px)
Citation preview
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
GROMACS inside TM
● Electrostatics● Force fields● Polarizability● Parallellization● Future technology
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics
● Coulomb's law gives the electrostatic interaction Vij between particles i and j at distance rij:
Vij=qi qj
40 rij
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics
● Long ranged, in particular with full charges, but even with large dipoles like in water.
● Cutoff methods give huge artefacts
Electrostatics
Yamada et al. Phys. Rev. Lett. 88 (2002) 195701
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics Reaction field
● Assume there is a dielectric continuum εRF
beyond the cutoff rc that modulates the
interaction between two charges. ● Interaction can then be written as:
VijRF=Vij1
RF−12RF1
rij3
rc3
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics – Reaction field
● Cheap to compute with short rc
● Reasonable results for dipolar liquids● Problematic for charged groups● Makes molecules more “ slippery” (i.e.,
higher mobility)
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics – Ewald
● Split the Coulomb interaction into two parts, a short range and a long range.
VijSR=Vij erfc
rij
VijLR=Vij[1−erfc
rij
]
Electrostatics – Ewald
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics – Ewald
● Compute the short range part of the potential directly
● Compute the long range part by solving the modified Poisson equation in Fourier space (for an infinitely periodic sample):
r=−r0
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics – Ewald
● Long range part can be done analytically (classical Ewald summation)
● Long range on a grid can be done using fast Fourier transform
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Electrostatics – Ewald
● Practical considerations:● Grid spacing● Cutoff distance● Distribution of charges on a grid using
splines (need to do the inverse when computing the forces).
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Force field status
● GROMOS and OPLS supported● Amber underway● CHARMM unknown?
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Force field status
● How to test force fields?● Organic molecules
Wensink et al. J. Chem. Phys. 119 (2003) 7308
Wensink et al. J. Chem. Phys. 119 (2003) 7308
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Force field status
● Organic molecule simulations teach us that force fields are qualitatively correct but not quantitatively yet.
● Combination of too large excess enthalpy and too low excess density may mean that potentials are too attractive overall and too repulsive at short distance
● Solution: get rid of the 12 term in Lennard Jones potential.
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Force field status
● How to test force fields?● Protein simulations (Van der Spoel &
Lindahl, J. Phys. Chem. B 107 (2003) 1117811187)
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Villin headpiece subdomain
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Villin headpiece subdomain
● 36 residue mini protein● NMR Structure known ● Thermostable● Hydrophobic core ● Protein stable under single mutations of the
core● Studied by by Duan and Kollman in a
single microsecond simulation
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Correspondence with NMR
● Compute distances derived from NOE data● Compare simulated to experimental results● The violation of experimental distances
(largest violation, number of violations and the sum of violations) determines the quality of the simulation
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
NMR distance restraints
● Sim. Nviol
(325) Maxviol
(nm) σviol
(nm)
● NMR 42 0.17 1.7● GROMOS 31 0.15 1.2● OPLS 42 0.22 2.0
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
NMR Chemical shifts
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Effect of starting structure
● OPLS● Max RMSD● 50 ns
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Force field status
● Simulation ensemble comparable to NMR ensemble, but not perfect compared to NMR data
● GROMOS slightly better than OPLS for Villin
● For Trpcage it was the other way around (not shown)
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Polarizability
● Phase transferability● Interfaces● Inhomogeneous systems● Proteinmetal interactions
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Polarizability
Shell Water (SW)Van Maaren & Van der Spoel,
J. Phys. Chem. B 105 (2001) 2618
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Polarizability
● k rSD
= qS E
● µind
= E
● k = qS
2/
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Polarizability – water dimer
● Model ROO
Epot
● SW 0.295 21.78 2.59● Dang 0.287 19.62● MCDHO 0.292 20.9 2.68● TIP4P 0.274 26.35 2.70● SPC/E 0.274 30.1 3.76● Ab Initio 0.291 21.0● Exper. 0.295 22.6(3) 2.60
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Polarizability
● Further work in progress, based on ab initio calculations for HF, Water, NH3, CH4.
● Automatic parameterisation tools available in GROMACS
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Parallellization
● Particle decomposition (particles are allocated to processors, which is simple algorithmically)
● In principle SIMD● Gromacs runs quite efficient with a cut
off, if the system is large enough● Not with PME
Scaling benchmarks
Scaling benchmarks
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Future parallellization
● Domain decomposition (chunks of space are given to each processors, more complex to program)
● PME on a few dedicated processors: real MIMD
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Future technology
● XML for file formats
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Why use XML?● Growing complexity of input files● Self contained file format which
can/could be validated● Separate data from algorithms (anti object
oriented!)● Extensible?● Machine readable (using e.g. libxml2)
Model
Other things
Residue -force fieldmapping
Non bondedatom types
Force field
Non bondedatom types
Bondedparameters
Non bondedparameters
Bondedatom types
Definitions
Simulationparameters
Integrationalgorithm
Outputcontrol
T, Pcontrol
Moreparameters
Links
Residues
Modifications
Moleculedescription
Modificationdefinitions
Linkdefinitions
Residuedefinitions
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Is this the right approach?
● DTD versus Schema● Storing information in attributes● Transformations using XSLT?● Use CML?
Model
Other things
Residue -force fieldmapping
Non bondedatom types
Force field
Non bondedatom types
Bondedparameters
Non bondedparameters
Bondedatom types
Definitions
Simulationparameters
Integrationalgorithm
Outputcontrol
T, Pcontrol
Moreparameters
Links
Residues
Modifications
Moleculedescription
Modificationdefinitions
Linkdefinitions
Residuedefinitions
By XSLT?
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Why use Python?● Easy to program● Portable● Can be used for GUI design using several
underlying libraries (wxWindows, FLTK, PyGTK, QtPython etc.)
● Can be used as a generic scripting language (rather than a dedicated scripting language)
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Python in GROMACS...
● In the first hand for developing a GUI● Using tools/libraries in a simple manner
(NetCDF/XML)● Interfacing with other programs (MMTK,
pymol, VMD)
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Python in GROMACS...
● Use the GROMACS library function parse_common_args to generate python scripts
● A simple generic dialog box routine using Tkinter
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Analysis front-end
David van der Spoel
Molecular Biophysics Group, Department of
Cell and Molecular Biology
spoel at xray.bmc.uu.se
CSC WorkshopEspoo
35 feb 2004
Acknowledgements Berk Hess (Groningen) Erik Lindahl (Stanford) Anton Feenstra (Amsterdam) Herman Berendsen (Groningen) Konrad Pywowarczik (Krakow) Alexandra Patriksson, Michiel van Lun
(Uppsala)